Discovery can start without sensitive data
For the first audit, you can describe the workflow or provide fake, anonymized, or sample files with the same column structure. We do not need real customer, bank, or transaction data to understand the initial process.
Default first-step data boundary
- Fake, anonymized, or sample data is preferred for the first review.
- Do not submit passwords, bank login credentials, identity documents, payment card numbers, API keys, or unredacted personal data through the website form.
- Use a description of the workflow if you are not sure whether a file is safe to share.
- Production file handling is agreed separately before implementation begins.
Human review stays in the workflow
For finance and operations reconciliation, we recommend automated detection plus human review. The workflow highlights exceptions; your team makes final business decisions.
Data access and retention
- Access should be limited to the files and fields required for the agreed workflow.
- Files used for implementation should be shared through agreed channels only.
- Retention and deletion expectations are confirmed before real production files are used.
- We can discuss NDAs and client-specific handling requirements before implementation.
Third-party tools and models
If a workflow requires OCR, document extraction, AI models, automation platforms, cloud storage, or other third-party services, those tools should be identified and agreed before production data is processed.
Client environment options
Where appropriate, workflows can be designed to run around your existing tools or inside a client-controlled environment. The best setup depends on file sources, security requirements, and operational constraints.
PDFs and scanned documents
For scanned documents or low-quality PDFs, extraction accuracy can vary. We may include validation and manual review steps instead of promising fully automatic classification.